中文

BlockServe: Block-Grained Continuous Batching for High-Throughput Diffusion LLM Serving

机器学习 2026-07-09 v1

摘要

Efficient serving of diffusion large language models (dLLMs) is hindered by convergence heterogeneity: when batching multiple requests, different sequences converge at different rates, causing faster requests to stall behind slower stragglers and introducing compute bubbles and tail latency. We present BlockServe, a continuous batching framework that integrates block-grained scheduling -- immediately evicting completed requests at block boundaries -- with mixed-state execution that extends dual cache and parallel decoding to heterogeneous batches via gather-scatter indexing. Furthermore, a compute-aware admission controller expands effective batch capacity through token-budgeted refill. On Dream and LLaDA across five benchmarks, BlockServe achieves 1.9--10.6×\times throughput over Fast-dLLM with comparable generation quality, establishing block-grained scheduling as a foundation for high-throughput offline dLLM inference.

引用

@article{arxiv.2607.08930,
  title  = {BlockServe: Block-Grained Continuous Batching for High-Throughput Diffusion LLM Serving},
  author = {Yuanjie Zhu and Liangwei Yang and Ke Xu and Weizhi Zhang and Shanghao Li and Zihe Song and Philip S. Yu},
  journal= {arXiv preprint arXiv:2607.08930},
  year   = {2026}
}